Leads, data and signals
How to tell whether a buying signal worked: measuring reply rate and meetings by signal
By Viktorio Shopov, co-founder of Sellinger. Reviewed by Asen Levov. Published . Last updated . 8 minute read.
To judge a buying signal, keep each signal source in its own campaign, write the same kind of message to every group, and compare meetings per 100 people messaged, not replies alone. Replies are noisy at small sizes. With about 100 people per group, a reply rate gap under roughly 10 points can be chance, and meeting rates need thousands. Our own account saw a tenfold spread between signal sources.
Buying signal tools say what they detect. Few say whether acting on it paid off. This post gives a method you can run in a week with data you already have, shows what it found on our own account, and says honestly where it breaks. It does not claim a result for any single signal template, because we have none we can publish.
What do you compare?
Three numbers per signal source, always with the same denominator, which is people messaged:
- Reply rate. People who replied, divided by people messaged. A human reply counts, refusals included, and automatic replies do not. Our post on how we define reply rate sets out the rules.
- Meetings per 100 messaged. Meetings booked divided by people messaged, times 100. This is the number that pays.
- Held meetings. Booked is a calendar entry. Held is that both people showed up. Count them separately, as in count meetings honestly.
A high reply rate with no meetings is a warning, not a win. It often means friendly replies from people who cannot buy.
How do you set the test up?
- One source per campaign. If funding announcements and post engagers share a campaign, you cannot separate them afterwards. In Sellinger, each signal lands leads in a segment you choose, so give each signal its own segment and campaign. Analytics can then be filtered by campaign.
- Same agent, same offer, same weeks. A signal tested in March against a list tested in June tells you about March and June.
- Same kind of opener. If one group gets an observation about the post and the other gets a generic line, you are testing the opener, not the signal. Our case study found observation openers out-replied questions by 10 to 15 points within the same month.
- Same stage. Compare people at the same step, for example after the first message.
- Write down the rule first. Decide before you look which number decides the test.
How many people do you need?
Fewer than most people use, and more than most people think. A rough rule from basic statistics, shown with invented numbers to make it concrete:
| Group size | Typical rate | Gap you can trust (about) |
|---|---|---|
| 100 people each | Reply rate near 20% | About 11 points |
| 300 people each | Reply rate near 20% | About 6 points |
| 1,000 people each | Meeting rate near 3% | About 1.5 points |
These come from the usual margin of error for the difference of two proportions at 95% confidence. They are our own arithmetic, not a measured result. The point: if one signal got 25 replies from 100 people and another got 18 from 100, you have not found a winner. If one got 54 from 100 and another 22 from 100, you probably have.
What did the test show on our own account?
On our own LinkedIn outbound, the Sellinger case study compares three sources. In the October 2025 cohort, with the same agents, intent signal lists replied at 54.6% and booked 4.54 meetings per 100 messaged, against 37.2% and 1.16 for an untagged list. A Sales Navigator search list over the whole period replied at 21.7% and booked 0.71.
The signal was engagement with a post, and the lesson is not "engagers win". Inside the signal group, people who had engaged with posts by AI SDR vendors and automation creators booked 2.6% to 3.7% per person messaged, and people who engaged with posts by enterprise sales thought leaders booked under 0.4%. The method was identical, and the audience whose posts were tracked made a tenfold difference. A signal is a rule for finding people. Whether those people can buy is a separate question.
These figures are one account, one offer and one audience. They show how to read the test, not what yours will show.
A worked example, with invented numbers
A founder who sells scheduling software for clinics tests two signals for three weeks. The numbers below are made up to show the steps.
| Source | People messaged | Replies | Reply rate | Meetings booked | Meetings per 100 |
|---|---|---|---|---|---|
| Posts asking for a scheduling tool | 140 | 49 | 35% | 6 | 4.3 |
| Clinics hiring a front desk role | 160 | 40 | 25% | 3 | 1.9 |
| Control: same job titles, no signal | 150 | 27 | 18% | 1 | 0.7 |
Reading it: the first source leads on both numbers, but with 6 and 3 meetings the meeting gap is within chance. The reply gap against the control, 35% against 18% with about 150 people each, is large enough to take seriously. The sensible next step is not to drop the second source. It is to send the winner and the runner up to 300 more people each and see whether the order holds.
Why do meetings take longer to read than replies?
Because they are rarer and arrive later. A reply comes within days. A meeting needs a reply, a yes and a free slot, so a group messaged this week may not show its meetings for two to three weeks. If you compare too early, the group with the longer path looks worse than it is. Wait for the tail, and say in your notes how long you waited. Count people once, however many times you wrote to them.
What goes wrong?
- Selection bias. People who engage with content may reply more regardless of any message. Compare to a list of equally targeted people with no signal, not to a random list.
- Different freshness. A signal acted on in two days against one acted on in three weeks. Record the age of the signal.
- Small samples. See the table above.
- Counting replies as results. Check what the replies lead to.
- Changing two things. New signal and new message in the same test.
- Chasing noise. Run the test again on a new group before you rebuild a campaign around a result.
How do you decide what to do next?
Use three outcomes. A signal that beats your baseline on meetings by a wide margin and holds on a second group gets more volume. One that matches your baseline can stay if it is cheaper to run. One that lags on meetings goes, even if it replies well. Our post on which signals are worth acting on gives a starting order to test, and lead scoring on LinkedIn shows how to rank people inside a list.
A one week plan
- Day 1: choose two signals and one control list. Create three segments and three campaigns.
- Day 2: write one opener style and use it for all three.
- Days 3 to 7: send in equal batches and read the first replies yourself.
- Week 3: count replies, meetings per 100 messaged and held meetings.
- Week 4: run the winner against a fresh group.
The guide to reading your analytics explains where each number sits in the app.
Common questions
How do I measure whether a buying signal works?
How many leads do I need to compare two signals?
Is a higher reply rate always better?
How do AI tools rank buying signals?
Can Sellinger report results by signal?
How Sellinger helps
Read how it works on See intent signals, or see a real campaign in Read how signals ran our own outbound.
